CN112288777A - Method for tracking laser breakpoint by using particle filtering algorithm - Google Patents
Method for tracking laser breakpoint by using particle filtering algorithm Download PDFInfo
- Publication number
- CN112288777A CN112288777A CN202011174850.5A CN202011174850A CN112288777A CN 112288777 A CN112288777 A CN 112288777A CN 202011174850 A CN202011174850 A CN 202011174850A CN 112288777 A CN112288777 A CN 112288777A
- Authority
- CN
- China
- Prior art keywords
- breakpoint
- particles
- state
- particle
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000002245 particle Substances 0.000 title claims abstract description 164
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 36
- 238000001914 filtration Methods 0.000 title claims abstract description 13
- 238000012952 Resampling Methods 0.000 claims abstract description 25
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000012544 monitoring process Methods 0.000 claims abstract description 18
- 230000007704 transition Effects 0.000 claims abstract description 17
- 238000005259 measurement Methods 0.000 claims abstract description 15
- 238000005295 random walk Methods 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000004927 fusion Effects 0.000 claims description 8
- 230000015556 catabolic process Effects 0.000 claims description 4
- 238000006731 degradation reaction Methods 0.000 claims description 4
- QVQLCTNNEUAWMS-UHFFFAOYSA-N barium oxide Chemical compound [Ba]=O QVQLCTNNEUAWMS-UHFFFAOYSA-N 0.000 claims description 3
- 229910001864 baryta Inorganic materials 0.000 claims description 3
- 238000011524 similarity measure Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 7
- 238000006073 displacement reaction Methods 0.000 description 6
- 238000003708 edge detection Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 239000003086 colorant Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000005653 Brownian motion process Effects 0.000 description 2
- 238000005537 brownian motion Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 241000270295 Serpentes Species 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000003760 hair shine Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000001678 irradiating effect Effects 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method for tracking laser breakpoints by using a particle filter algorithm, which comprises 6 steps of establishing a state space model, initializing targets and particles, measuring similarity, updating weights, determining breakpoint states and judging resampling; describing breakpoint monitoring by adopting a dynamic state space model, wherein the dynamic state space model of the time-varying problem comprises a state transition model and an observation model, and determining the dynamic state space model; initializing a breakpoint target and sample particles; updating the weight of the particles through similarity measurement, wherein each particle has different weights; based on a particle filtering algorithm, acquiring a particle update weight according to an observation likelihood function of the target region characteristics; and correcting the particles with different weights, performing recursive prediction and updating on the breakpoint state, and determining the breakpoint state. The invention can detect the laser breakpoint in the video screen in real time, thereby giving real-time early warning to the landslide, and the tracking process is accurate, stable and real-time.
Description
Technical Field
The invention is applied to a laser breakpoint tracking technology during real-time early warning of a slope surface, and particularly relates to a method for tracking a laser breakpoint by using a particle filtering algorithm.
Background
Landslide is one of the most serious geological disasters, China is seriously threatened by landslide disasters, and according to statistics, economic losses caused by landslide are 150-200 billion yuan of RMB.
Common technologies for monitoring landslide surface deformation are geodetic precision measurement, GPS method, photography, and the like. The earth precision measurement method adopts high-precision optical and photoelectric measuring instruments, such as a precision level gauge, a total station and the like, and completes the monitoring task through angle measurement and distance measurement. Its advantages are high absolute displacement of slope and high precision. But the disadvantages of the method are limited by topographic conditions and meteorological conditions, large workload, long period and poor continuous observation capability. The GPS method is used as a technical means of modern geodetic survey, can realize three-dimensional geodetic survey, and has the advantages of short observation time, all-weather real-time monitoring, no need of mutual communication among monitoring stations, small influence by the outside and the like. The GPS landslide monitoring network mainly comprises datum points and monitoring points. But when the position of the monitoring point is selected, the situation that a television station, a radio station, a mobile communication station, a microwave relay station, a transformer substation, a high-voltage wire, a sheet obstacle and the like cannot exist around the monitoring point is required to be ensured. The photography method is a method for measuring the three-dimensional coordinates of observation points of an image sheet by using a three-dimensional coordinate instrument when a camera is placed on 2 different fixed points and a three-dimensional image is formed by photographing the observation points in a slope range. Its advantages are convenient operation, saving time and labour, and the obtained photo data is the real-time record of landslide surface change. Its disadvantages are the low accuracy of the observation and the large influence of climatic conditions (e.g. rainy days, night, etc.).
The key of real-time early warning of landslide is to detect the laser breakpoint in the video in real time, and if tracking the laser breakpoint is the key of the monitoring process.
It is noted that this section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. The description herein is not admitted to be prior art by inclusion in this section.
Disclosure of Invention
The invention aims to provide a method for tracking laser breakpoints by using a particle filtering algorithm, which ensures the accuracy, stability and instantaneity of a tracking process.
In order to achieve the purpose, the invention adopts the following technical scheme:
the method for tracking the laser breakpoint by using the particle filter algorithm comprises the following steps:
s1: establishing a state space model
Describing breakpoint monitoring by adopting a dynamic state space model, wherein the dynamic state space model of the time-varying problem comprises a state transition model and an observation model, and determining the dynamic state space model;
s2: initializing targets and particles
Initializing a breakpoint target and sample particles;
s3: similarity measure
Updating the weight of the particles through similarity measurement, wherein each particle has different weights;
s4: weight value updating
Based on a particle filtering algorithm, acquiring a particle update weight according to an observation likelihood function of the target region characteristics;
s5: determination of breakpoint status
Correcting the particles with different weights, performing recursive prediction and updating on the breakpoint state, and determining the breakpoint state;
s6: resampling determination
And (3) resampling by adopting a particle filter algorithm, reducing the particles with smaller weight, and increasing the particles with larger weight, thereby correcting the posterior probability density.
Further, the step S1 includes the following steps:
s101: state transition model
Regarding continuous video image frames as a continuous time sequence at a corresponding moment, continuously updating the state of a breakpoint along with the lapse of time, and adopting an autoregressive model or a random walk model;
1) random walk model
The random walk model refers to the position of the target at the next time is unpredictable, and the state transition equation of the random walk model can be defined as:
xk＝xk-1+vk-1 (1)
wherein: x is the number ofkIs the state of the target at time k; v. ofk-1Is the process noise at time k-1;
2) autoregressive model
The autoregressive model establishes a regression equation to predict the target state according to the relation between the current state of the system and the historical state, and the motion model can be defined as follows:
xk＝a1xk-1+a2xk-2+…+anxk-n+vk-1 (2)
wherein: x is the number ofkIs the state of the target at time k; v. ofk-1Is the process noise at time k-1; a is1,a2,…,anIs a regression coefficient;
s102: observation model
Establishing an observation model by utilizing the characteristics of the tracked target, and correcting and updating the predicted position through the observation model; wherein the features of the tracked object comprise color features, edge features and texture features.
Further, the step S3 is specifically:
assume that the discrete probability distribution models of the reference object and the candidate object are p respectivelyuAnd q isuThen the Bhattacharyya coefficient can be calculated by:
wherein p isuAnd q isuAre normalized values; the value range of rho is 0-1;
bhattacharyya distance (baryta distance) calculation formula:
further, the step S4 is specifically:
the observed likelihood function of the target region color features is noted as:
the observed likelihood function of the edge feature of the target region is noted as:
the particle weights based on color and edge characteristics are respectively:
wherein k represents a frame number, and i represents an ith particle;
and adopting a linear fusion mode of the color features and the edge features, wherein the final weight value determined by the color features and the edge features is as follows:
normalizing the weight of the particles:
a multiplicative fusion strategy may also be employed, with the particle weights of the fused color and edge features being:
then, normalizing the weight:
further, the step S5 is specifically:
for the determination of the breakpoint state, the particles with different weights are used to correct the breakpoint state, and the following three criteria are specifically adopted to determine the final state of the breakpoint:
1) maximum a posteriori criterion
Taking the state represented by the particle with the maximum weight in the posterior particles as the final state of the breakpoint, specifically as follows:
2) global weighting criteria
Considering the contribution of each candidate particle to the state estimation, and taking the result obtained by weighting and summing all the particles as the final estimated state of the target;
3) local weighting criterion
Selecting M particles from a group within a certain radius range for weighting by taking the particle with the largest weight as a center, wherein the calculation formula is shown as the following formula, wherein M is less than or equal to N:
further, the step S6 is specifically:
the posterior probability density of a particle at time k is:
system posterior probability density after resampling
Can be expressed as:
wherein N isiRepresenting the particles x during resamplingiThe weight re-amplitude of each particle after resampling is as
The statistical rule of the resampling algorithm is the number of valid particles, which is defined as follows:
wherein N is the number of the particles,
the weight of the particles in the real state is represented, and the calculation formula is as follows:
wherein the content of the first and second substances,
the weight value of the normalization of the particles is represented,
indicating the condition of particle degradation.
The invention has the beneficial effects that:
1) the invention discloses a method for tracking laser breakpoint by using a particle filter algorithm, which realizes stable tracking by adopting a linear fusion mode of color characteristics and edge characteristics; the state of the breakpoint is determined by adopting a local weighting mode, so that noise is eliminated, and the result is smoother and more reliable.
2) The invention discloses a method for tracking a laser breakpoint by using a particle filter algorithm, which aims to monitor a landslide in real time, track the laser breakpoint by using the particle filter algorithm in real time, draw the displacement of the breakpoint and send out landslide early warning when the displacement of the breakpoint exceeds a preset threshold.
3) The invention discloses a method for tracking laser breakpoints by using a particle filtering algorithm, which has accurate, stable and real-time tracking process; the particle filter algorithm has the advantages of easy understanding, easy realization, few parameters, strong algorithm robustness and the like.
Drawings
FIG. 1 is a flow chart of the present invention
FIG. 2 is a tracking schematic of the present invention;
FIG. 3 is a dynamic state space model of the present invention;
fig. 4 is a schematic structural diagram of a video monitoring system according to the present invention.
In the figure: 1-slope surface; 2-a rigid baffle; 3-a camera; 4-a GSM communication module; 5-a red laser emitter; 6-an alarm; 7-a computer; 8-server.
Detailed Description
Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The described features or characteristics may be combined in any suitable manner in one or more embodiments.
The video monitoring system for landslide by using red laser as shown in fig. 4 comprises a plurality of rigid baffles 2, a camera 3 for collecting position changes of the rigid baffles 2, a GSM communication module 4 for transmitting monitoring information, a red laser transmitter 5 for transmitting red laser, an alarm 6 for warning and reminding, and a computer 7 for receiving and processing data.
The rigid baffle 2 is distributed on the monitored slope surface 1; the red laser emitter 5 is arranged opposite to the slope surface 1, and red laser emitted by the red laser emitter 5 is opposite to the rigid baffle plate 2; the camera 3 is positioned opposite to the slope surface 1 and is opposite to the rigid baffle 2; the GSM communication module 4 is connected with the camera 3, and the GSM communication module 4 is connected with the computer 7. The system also comprises alarm devices 6 and 8, wherein the alarm device 6 is connected with the computer 7, and the server 8 is connected with the computer 7. The GSM communication 4 and the computer 7 are connected wirelessly and remotely.
When red laser emitter 5 shines rigid baffle 2 of installing on domatic 1, because rigid baffle 2 can shelter from laser, the breakpoint can appear in the red laser line that leads to originally complete. According to the principle, rigid baffles 2 with the same size are respectively arranged at the points of the slope surface needing to be monitored in an important mode, and red laser is used for irradiating the rigid baffles 2 respectively. We use the camera 3 to capture the view screens of each laser breakpoint on the rigid barrier 2 and send these views to the computer 7 for processing via the GSM communication module 4. After receiving the data, the computer 7 uses a particle filtering algorithm to detect the laser breakpoints in real time, draws the motion tracks of the breakpoints, realizes the real-time tracking of the laser breakpoints, and judges whether to send out landslide early warning according to the displacement of the laser breakpoints.
The key of the technology whether the landslide can be early warned in real time is to detect the laser breakpoint in the video in real time. The process of tracking laser breakpoints by using a particle filtering algorithm is roughly divided into 6 steps: establishing a state space model, initializing targets and particles, measuring similarity, updating weights, determining breakpoint states and judging resampling.
a. Establishing a state space model
And describing the breakpoint monitoring problem by adopting a dynamic state space model. The dynamic state space model of the time-varying problem is divided into: a state transition model and an observation model. The state transition model reflects a target state transition mode, and the observation model describes target characteristics.
(1) State transition model
Based on video landslide monitoring, laser breakpoints are targets to be tracked. Successive frames of video images may be viewed as a continuous time sequence of corresponding time instants. The state of the breakpoint is constantly updated over time.
The main purpose of breakpoint tracking in the present invention is to find the accurate position of the laser breakpoint in each frame, so the state information of the target breakpoint is mainly the pixel coordinate position of the laser breakpoint in the video frame.
The breakpoint occurs at a different position in each frame, and the position status is updated over time, which we can consider as a motion model. In an actual scene, the set motion model not only conforms to the breakpoint motion rule, but also is convenient for computer operation processing, so that a perfect model is difficult to find in reality, and most of the set motion model is approximated by some common motion models. Common models used in the field of visual tracking are: autoregressive model, random walk model.
1) Random walk model
The random walk model is also called random walk, and the position of the target at the next moment is unpredictable, is close to brownian motion, and is an ideal mathematical state of the brownian motion. The state transition process of the random walk model can be defined as:
xk＝xk-1+vk-1 (1)
wherein: x is the number ofkIs the state of the target at time k; v. ofk-1Is the process noise at time k-1.
2) Autoregressive model
The autoregressive model establishes a regression equation to predict the target state according to the relation between the current state of the system and the historical state, and the motion model can be defined as follows:
xk＝a1xk-1+a2xk-2+…+anxk-n+vk-1 (2)
wherein: x is the number ofkIs the state of the target at time k; v. ofk-1Is the process noise at time k-1; a is1,a2,…,anAre regression coefficients.
If the transition process of the target state is regarded as a randomly varying process, the parameters in the above equation satisfy the constraint condition: a is1＝1，ai0 (i-2, 3, …, n). At this time, the autoregressive model is a first-order autoregressive model, and the autoregressive model degenerates to a random walk model, so that the random walk model is a special case of the autoregressive model.
After the state transition model is established, the representing breakpoint moves according to the motion model, namely, the propagation mode of the particle is determined. The position of the breakpoint can be predicted through the model, and then the predicted position is corrected and updated through the observation model.
(2) Observation model
Observing the effect of the model: and correcting the real posterior probability density by using the observation value. The breakpoint tracking problem related to the invention mainly refers to: extracting the image characteristics of the region where each particle is located at the moment k, and then comparing the image characteristics with the characteristics of a target breakpoint (target template) to obtain a weighing value of a similar degree, namely the weight of the particle; then, the predicted state of the breakpoint is updated according to the value. The process of updating the weight of the particle is the process of measuring the similarity, the particle similar to the real state of the target breakpoint is distributed with a larger weight, and the particle deviating from the real state with a larger weight is distributed with a smaller weight.
In the field of target tracking, it is most common to establish an observation model by using the characteristics of a tracked target. The tracking features are mainly: color features, edge features, texture features, and the like.
The purpose of selecting features is to distinguish them from other things, the more unique features are advantageous in the tracking process. The color characteristic is the most visual attribute of the moving object, the edge characteristic can be clearly distinguished from the background, and the texture characteristic can reflect the detail information of the object. The selection of features may affect the effectiveness of the tracking.
When the method is used for monitoring the landslide, the selection of the image characteristics is particularly important, and the quality of the tracking performance is often in great relation with the selection of the image characteristics. The selection of proper characteristics has great influence on the accuracy, stability and real-time performance of the algorithm.
1) Color characteristics
The color characteristics can express the characteristics of the target area most intuitively, and the method has the characteristics of good stability and insensitivity to direction. Color features can be expressed using a variety of color space models, with common color spaces being RGB, YCbCr, HSV, HSL, and the like. The RGB, YCbCr and other spaces are related to hardware equipment, and the numerical values of the corresponding channels can be generally directly acquired from an original image; the two color spaces HSV and HSL are oriented to users, are more consistent with the visual perception of human color, and respectively describe three typical color components quantitatively. It is often necessary to convert the color space of the object from device-oriented to user-oriented. In the field of computer video image processing, colors are often used: an RGB color space model, an HSV color space model.
The method for representing the color features mainly comprises the following steps: color histograms, color moments, color aggregate vectors, and the like. The color histogram has the advantages of simple extraction, rotational invariance, insensitivity to local shielding and the like, so the color histogram is a common color feature representation method. Color histograms will also be used herein to represent color characteristics.
2) Edge feature
The edge features reflect the contour information of the object and also include information such as direction and step property. If the observation model is built with shape features, then the image features such as corners, edges, contours, etc. will be of interest. The edge features have excellent characteristics that are insensitive to illumination and color variations.
Common algorithms are: corner detection operators, such as: moravec operator, Harris operator, etc.; gradient operators for angular edge features, such as: sobel operator, Laplacian operator, Canny edge detection operator, etc.; contour detection operators, such as: snake contour extraction operators, level set contour extraction operators and the like.
The invention adopts Prewitt edge detection operator to extract the edge characteristics of the target breakpoint in the video image. The Prewitt edge detection operator can better detect and track the target edge and can effectively inhibit noise in the video image.
And after the edge features are extracted, establishing a target observation model by utilizing the edge direction histogram. Suppose that the image target region Φ (x) has a pixel point R (Φ) corresponding theretox) Calculating the first partial derivatives of the pixel points in the x direction and the y direction respectively to obtain the gradient value R in the horizontal directionxAnd a vertical gradient value Ry. Then, the expressions of the amplitude f and the direction angle of the edge gradient of the pixel point are respectively:
the u-th edge direction histogram feature in the target region is expressed as:
b initializing targets and particles
After the spatial model is determined, the breakpoint status can be recursively predicted and updated to determine the breakpoint status. But before the recursive operation we need to perform initialization operations on the breakpoint target and the sample particle.
Initializing a target breakpoint and a sample particle:
(1) breakpoint initialization
First, the region where the target breakpoint is located is selected. Selecting a rectangular frame with a break point at a target in a first frame in a manually specified mode;
then, extracting and recording the characteristics of the break points in the rectangular frame area for the similarity measurement process; (can extract the color characteristic and shape characteristic of the breakpoint, calculate the color histogram and shape histogram of the target breakpoint, and build the target model)
After the breakpoint characteristic information is extracted, the state information of the breakpoint is also required to be acquired. For breakpoint tracking, its state refers to the location coordinates. In order to improve the tracking accuracy, the length and width information of the breakpoint region is added as part of the state information.
(2) Particle initialization
The state of each particle is used as a candidate state of the breakpoint and needs to be changed continuously with the lapse of time, and when the number of the particles is larger, the candidate state of the breakpoint is richer, and the finally obtained breakpoint state is more accurate.
For the initialization of the particles, only the state information thereof needs to be initialized. In the initial frame, a large number of particles are scattered to be distributed around the target break point, namely the initialization of the particles.
It is common practice to distribute the scattered particles uniformly in the rectangular area where the break points are located, and the size of each particle is consistent with the initialized break point size. At the initial moment, because each particle is considered equally important, the weight of each particle is initialized to 1/N, where N is the number of particles.
c similarity measure
And updating the particle weight value through the similarity measurement.
Regardless of the type of feature used, after finding the features of the candidate target region, attention is paid to the similarity measurement with the original target, and a corresponding relation is established to calculate the probability that the region is the target. For the target characteristics with numerical values, similarity measurement can be carried out through Euclidean distance and Manhattan distance; for features whose results are described in the form of a histogram,then the method can be implemented by: chi shape2Distance, Bhattacharyya distance (papanicolaou distance) is used for similarity measurement. Finally, the observed likelihood function may be defined by the probability of the target being proportional to the similarity of the features.
Since the invention represents features in the form of histograms, the Bhattacharyya distance is used for similarity measurement.
Assume that the discrete probability distribution models of the reference object and the candidate object are p respectivelyuAnd q isuThen the Bhattacharyya coefficient can be calculated by:
wherein p isuAnd q isuAre normalized values; the value range of rho is 0-1, the larger the value of rho is, the more similar the two probability models are, and when rho is 1, the two models are completely matched.
Bhattacharyya distance (baryta distance) calculation formula:
according to the formula, the similarity d between the histogram model of the candidate target area and the histogram model of the original target area of the edge feature can be respectively calculatededgeSimilarity d between candidate target region histogram model of color features and original target region histogram modelcolor。
d weight updating
In the particle filtering algorithm, the Bhattacharyya distance (babbit distance) cannot be directly used as the particle weight, and the particle update weight is obtained according to the observation likelihood function of the target region feature.
The observed likelihood function of the target region color features is noted as:
the observed likelihood function of the edge feature of the target region is noted as:
the particle weights based on color and edge characteristics are respectively:
where k denotes a frame number and i denotes an ith particle.
The method adopts a linear fusion mode of color features and edge features to realize stable tracking. Therefore, the final weight determined by the color feature and the edge feature is:
normalizing the weight of the particles:
besides linear fusion, a multiplicative fusion strategy can be adopted, and the weight of the particles fusing the color and the edge features is as follows:
then, normalizing the weight:
determination of e-breakpoint status
After similarity measurement is carried out on each candidate particle according to the prior characteristics, the importance among the particles is not equal any more, and each particle has different weight values. The closer to the particle of the breakpoint real state, the larger the weight, and conversely, the smaller the weight. For the final determination of the breakpoint status, it is a process of correcting the breakpoint status by using these particles with different weights, and there are generally three criteria for determining the final status of the breakpoint:
(1) maximum a posteriori criterion
And taking the state represented by the particle with the maximum weight in the posterior particles as the final state of the breakpoint, wherein the higher the similarity is, the more the real state of the target can be represented.
(2) Global weighting criteria
Considering the contribution of each candidate particle to the state estimation, the weight value of each candidate particle is the amount of the contribution, and the result of the weighted sum of all the particles is used as the final estimated state of the target.
(3) Local weighting criterion
And selecting M particles from a group within a certain radius range for weighting by taking the particle with the largest weight as a center, which is equivalent to performing windowing. The calculation formula is shown as the following formula, wherein M is less than or equal to N, and the equal sign is equivalent to the integral weight.
If the weight of the particles is unimodal distribution, the maximum posterior criterion is proper, but is greatly influenced by noise; if the weight distribution of the particles is multimodal, the weight of more particles is larger, the maximum posterior criterion can not accurately reflect the real state of the breakpoint, and the influence of noise is obvious. Whether multimodal or unimodal, the weighting criterion is more appropriate due to its smoothness.
Therefore, a locally weighted approach will be employed herein to determine the state of the breakpoint. Firstly, calculating the weight average value of all particles, and then carrying out weighted summation on the particle states with the weights larger than the average value, thereby obtaining the state of the breakpoint. This both eliminates noise and makes the result smoother and more reliable.
f resampling algorithm
The resampling of the particle filter algorithm is mainly to reduce the particles with smaller weight and increase the particles with larger weight, so as to correct the posterior probability density.
The posterior probability density of a particle at time k is:
then, the system posterior probability density after resampling
Can be expressed as:
wherein N isiRepresenting the particles x during resamplingiThe weight re-amplitude value of each particle after resampling is as
The process of particle resampling is to reselect the particles to form a new particle set, and if the weight is larger, the particles are copied for a plurality of times, and if the weight is smaller, the particles are copied for a plurality of times.
Resampling algorithms usually contain some decision criteria, where one common statistical rule is the number of valid particles, defined as follows:
wherein N is the number of the particles,
representing the weight of the particle in the true state is generally difficult to obtain. Therefore, many documents in the calculation use approximate calculation formulas:
wherein the content of the first and second substances,
representing the normalized weight of the particle.
The invention aims to monitor landslide in real time. The method is characterized in that a particle filter algorithm is used for tracking laser breakpoints in real time, the displacement of the breakpoints is drawn, and when the displacement of the breakpoints exceeds a preset threshold value, landslide early warning is sent out.
In combination with the above aspects, the simple flow of the particle filter algorithm is as follows:
step 1: initialization (k is 0)
1) And (5) specifying the initial state of the breakpoint, generating a corresponding sample, and extracting a breakpoint characteristic subspace.
2) The initial set of particles is sampled uniformly according to the system state prior distribution, i.e.: the particles were scattered evenly around the target while being given the same weight 1/N.
Step 2: generating N particles using a state transition model;
and (4) carrying out particle propagation on the particles according to the state transition model to form a new particle set. All particle children represent the possible states of the breakpoint, and the greater the number of particles, the more possible states are included. (Generation of a New set of particles based on the State transition model)
And step 3: calculating corresponding color and edge histogram models based on the observation models, and then performing similarity measurement; then, updating the weight of the particles and normalizing the weight;
the method comprises the following steps: firstly, converting a video image from an RGB color space to an HSV space, and establishing a color histogram model in the HSV space; then, converting the video image from HSV color space to a gray image, denoising the gray image by using a filtering method, extracting edge characteristics by using an edge detection algorithm, and establishing an edge histogram model; calculating color and edge histogram models of the target breakpoint region and the candidate target breakpoint region, and performing similarity measurement by using the Pasteur distance formula to obtain color similarity d between the target region and the candidate regioncolorEdge similarity dedge(ii) a Finally, calculating the observation likelihood functions of the colors and the edge characteristics of the target area according to the similarity values, and respectively calculating the weights of the colors and the edge characteristics so as to obtain the final weight of the particles and normalizing the weights;
and 4, step 4: estimating a target breakpoint state;
and 5: updating a target template;
and determining whether the target template needs to be updated according to information such as reconstruction errors. Model updates do not have a uniform criterion and generally consider the appearance of the target to change continuously, so the model is often updated every frame. However, it is also considered that the past appearance of the target is important for tracking, and continuous updating may lose past appearance information and introduce excessive noise, so that the problem is solved by combining long and short-term updating.
Step 6: judging whether to perform resampling;
and judging whether to perform resampling according to the effective sample scale. The threshold for the effective sample size is typically set to two-thirds of the number of particles, i.e. the
When the number of effective samples is less than NthrThen re-sampling is carried out; otherwise, no resampling is carried out, and the step 7 is entered.
And 7: determine if there is a next frame? If yes, according to the conditions of the particles, the step 2 is carried out again to carry out circulation; if not, exit.
Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
Claims (6)
1. A method for tracking laser breakpoints by using a particle filter algorithm is characterized by comprising the following steps:
s1: establishing a state space model
Describing breakpoint monitoring by adopting a dynamic state space model, wherein the dynamic state space model of the time-varying problem comprises a state transition model and an observation model, and determining the dynamic state space model;
s2: initializing targets and particles
Initializing a breakpoint target and sample particles;
s3: similarity measure
Updating the weight of the particles through similarity measurement, wherein each particle has different weights;
s4: weight value updating
Based on a particle filtering algorithm, acquiring a particle update weight according to an observation likelihood function of the target region characteristics;
s5: determination of breakpoint status
Correcting the particles with different weights, performing recursive prediction and updating on the breakpoint state, and determining the breakpoint state;
s6: resampling determination
And (3) resampling by adopting a particle filter algorithm, reducing the particles with smaller weight, and increasing the particles with larger weight, thereby correcting the posterior probability density.
2. The method for tracking laser breakpoint using particle filter algorithm according to claim 1, wherein the step S1 includes the steps of:
s101: state transition model
Regarding continuous video image frames as a continuous time sequence at a corresponding moment, continuously updating the state of a breakpoint along with the lapse of time, and adopting an autoregressive model or a random walk model;
1) random walk model
The random walk model refers to the position of the target at the next time is unpredictable, and the state transition equation of the random walk model can be defined as:
xk＝xk-1+vk-1 (1)
wherein: x is the number ofkIs the state of the target at time k; v. ofk-1Is the process noise at time k-1;
2) autoregressive model
The autoregressive model establishes a regression equation to predict the target state according to the relation between the current state of the system and the historical state, and the motion model can be defined as follows:
xk＝a1xk-1+a2xk-2+…+anxk-n+vk-1 (2)
wherein: x is the number ofkIs the state of the target at time k; v. ofk-1Is the process noise at time k-1; a is1,a2,…,anIs a regression coefficient;
s102: observation model
Establishing an observation model by utilizing the characteristics of the tracked target, and correcting and updating the predicted position through the observation model; wherein the features of the tracked object comprise color features, edge features and texture features.
3. The method for tracking laser breakpoints by using a particle filter algorithm according to claim 2, wherein the step S3 specifically comprises:
assume that the discrete probability distribution models of the reference object and the candidate object are p respectivelyuAnd q isuThen the Bhattacharyya coefficient can be calculated by:
wherein p isuAnd q isuAre normalized values; the value range of rho is 0-1;
bhattacharyya distance (baryta distance) calculation formula:
4. the method for tracking laser breakpoints by using a particle filter algorithm according to claim 3, wherein the step S4 specifically comprises:
the observed likelihood function of the target region color features is noted as:
the observed likelihood function of the edge feature of the target region is noted as:
the particle weights based on color and edge characteristics are respectively:
wherein k represents a frame number, and i represents an ith particle;
and adopting a linear fusion mode of the color features and the edge features, wherein the final weight value determined by the color features and the edge features is as follows:
normalizing the weight of the particles:
a multiplicative fusion strategy may also be employed, with the particle weights of the fused color and edge features being:
then, normalizing the weight:
5. the method for tracking laser breakpoints by using a particle filtering algorithm according to claim 4, wherein the step S5 specifically comprises:
for the determination of the breakpoint state, the particles with different weights are used to correct the breakpoint state, and the following three criteria are specifically adopted to determine the final state of the breakpoint:
1) maximum a posteriori criterion
Taking the state represented by the particle with the maximum weight in the posterior particles as the final state of the breakpoint, specifically as follows:
2) global weighting criteria
Considering the contribution of each candidate particle to the state estimation, and taking the result obtained by weighting and summing all the particles as the final estimated state of the target;
3) local weighting criterion
Selecting M particles from a group within a certain radius range for weighting by taking the particle with the largest weight as a center, wherein the calculation formula is shown as the following formula, wherein M is less than or equal to N:
6. the method for tracking laser breakpoints by using a particle filter algorithm according to claim 5, wherein the step S6 specifically comprises:
the posterior probability density of a particle at time k is:
system posterior probability density after resampling
Can be expressed as:
wherein N isiRepresenting the particles x during resamplingiThe weight re-amplitude of each particle after resampling is as
The statistical rule of the resampling algorithm is the number of valid particles, which is defined as follows:
wherein N is the number of the particles,
the weight of the particles in the real state is represented, and the calculation formula is as follows:
wherein the content of the first and second substances,
the weight value of the normalization of the particles is represented,
indicating the condition of particle degradation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011174850.5A CN112288777A (en) | 2020-12-16 | 2020-12-16 | Method for tracking laser breakpoint by using particle filtering algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011174850.5A CN112288777A (en) | 2020-12-16 | 2020-12-16 | Method for tracking laser breakpoint by using particle filtering algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112288777A true CN112288777A (en) | 2021-01-29 |
Family
ID=74372388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011174850.5A Pending CN112288777A (en) | 2020-12-16 | 2020-12-16 | Method for tracking laser breakpoint by using particle filtering algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112288777A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20140052256A (en) * | 2012-10-24 | 2014-05-07 | 계명대학교 산학협력단 | Real-time object tracking method in moving camera by using particle filter |
| CN103985139A (en) * | 2014-05-20 | 2014-08-13 | 重庆大学 | Particle filter target tracking method based on color model and prediction vector cluster model information fusion |
| CN104616318A (en) * | 2015-01-22 | 2015-05-13 | 重庆邮电大学 | Moving object tracking method in video sequence image |
| CN104680557A (en) * | 2015-03-10 | 2015-06-03 | 重庆邮电大学 | Intelligent detection method for abnormal behavior in video sequence image |
| CN105405151A (en) * | 2015-10-26 | 2016-03-16 | 西安电子科技大学 | Anti-occlusion target tracking method based on particle filtering and weighting Surf |
| CN106780560A (en) * | 2016-12-29 | 2017-05-31 | 北京理工大学 | A kind of feature based merges the bionic machine fish visual tracking method of particle filter |
| CN107563342A (en) * | 2017-09-15 | 2018-01-09 | 武汉大学 | A kind of pedestrian's robust tracking method searched and rescued towards unmanned plane field |
| CN108802692A (en) * | 2018-05-25 | 2018-11-13 | 哈尔滨工程大学 | A kind of method for tracking target based on maximum cross-correlation entropy volume particle filter |
| CN110363113A (en) * | 2019-06-28 | 2019-10-22 | 华南理工大学 | VLC-ITS information source based on particle filter tracks extracting method |
| CN111369597A (en) * | 2020-03-09 | 2020-07-03 | 南京理工大学 | Particle filter target tracking method based on multi-feature fusion |
-
2020
- 2020-12-16 CN CN202011174850.5A patent/CN112288777A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20140052256A (en) * | 2012-10-24 | 2014-05-07 | 계명대학교 산학협력단 | Real-time object tracking method in moving camera by using particle filter |
| CN103985139A (en) * | 2014-05-20 | 2014-08-13 | 重庆大学 | Particle filter target tracking method based on color model and prediction vector cluster model information fusion |
| CN104616318A (en) * | 2015-01-22 | 2015-05-13 | 重庆邮电大学 | Moving object tracking method in video sequence image |
| CN104680557A (en) * | 2015-03-10 | 2015-06-03 | 重庆邮电大学 | Intelligent detection method for abnormal behavior in video sequence image |
| CN105405151A (en) * | 2015-10-26 | 2016-03-16 | 西安电子科技大学 | Anti-occlusion target tracking method based on particle filtering and weighting Surf |
| CN106780560A (en) * | 2016-12-29 | 2017-05-31 | 北京理工大学 | A kind of feature based merges the bionic machine fish visual tracking method of particle filter |
| CN107563342A (en) * | 2017-09-15 | 2018-01-09 | 武汉大学 | A kind of pedestrian's robust tracking method searched and rescued towards unmanned plane field |
| CN108802692A (en) * | 2018-05-25 | 2018-11-13 | 哈尔滨工程大学 | A kind of method for tracking target based on maximum cross-correlation entropy volume particle filter |
| CN110363113A (en) * | 2019-06-28 | 2019-10-22 | 华南理工大学 | VLC-ITS information source based on particle filter tracks extracting method |
| CN111369597A (en) * | 2020-03-09 | 2020-07-03 | 南京理工大学 | Particle filter target tracking method based on multi-feature fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9495764B1 (en) | Verifying object measurements determined from mobile device images | |
| US10403037B1 (en) | Verifying object measurements determined from mobile device images | |
| CN109598794B (en) | Construction method of three-dimensional GIS dynamic model | |
| Rout | A survey on object detection and tracking algorithms | |
| CN111144213B (en) | Object detection method and related equipment | |
| KR101787542B1 (en) | Estimation system and method of slope stability using 3d model and soil classification | |
| CN108804992B (en) | Crowd counting method based on deep learning | |
| CN105279772B (en) | A kind of trackability method of discrimination of infrared sequence image | |
| CN101167086A (en) | Human detection and tracking for security applications | |
| Nyaruhuma et al. | Verification of 2D building outlines using oblique airborne images | |
| Kang et al. | The change detection of building models using epochs of terrestrial point clouds | |
| WO2009039350A1 (en) | System and method for estimating characteristics of persons or things | |
| US20220128358A1 (en) | Smart Sensor Based System and Method for Automatic Measurement of Water Level and Water Flow Velocity and Prediction | |
| CN112183434B (en) | Building change detection method and device | |
| CN112541938A (en) | Pedestrian speed measuring method, system, medium and computing device | |
| CN112085778A (en) | Oblique photography illegal building detection method and system based on superpixels and morphology | |
| CN115222884A (en) | Space object analysis and modeling optimization method based on artificial intelligence | |
| CN110851978A (en) | Camera position optimization method based on visibility | |
| CN110636248A (en) | Target tracking method and device | |
| CN103688289A (en) | Method and system for estimating a similarity between two binary images | |
| WO2022045877A1 (en) | A system and method for identifying occupancy of parking lots | |
| JP7243372B2 (en) | Object tracking device and object tracking method | |
| Liu et al. | Video monitoring of Landslide based on background subtraction with Gaussian mixture model algorithm | |
| Pless | Spatio-temporal background models for outdoor surveillance | |
| JP2002074369A (en) | System and method for monitoring based on moving image and computer readable recording medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |